基于卷积神经网络多层特征提取的目标识别

目标识别一直是人工智能领域的热点问题. 为了提高目标识别的效率，提出了基于卷积神经网络多层特征提取的目标识别方法. 该方法将图像输入卷积神经网络进行训练，在网络的每个全连接层分别进行特征提取，将得到的特征依次输入到分类器，对输出结果进行比较. 选取经过修正线性单元relu函数激活的低层全连接层作为特征提取层，比选取高层全连接层特征提取的识别率高. 本文构建了办公用品数据集，实现了基于卷积神经网络多层特征提取的办公用品识别系统. 选择AlexNet卷积神经网络模型的relu6层作为特征选取层，选择最优训练图像数量和最优分类器构建系统，从而证明了该方法的可行性.

Target Recognition Based on Multilayer Feature Extraction of Convolution Neural Network

Target recognition has been the hot issue in the field of artificial intelligence. In order to enhance the efficiency of target recognition, this paper proposes a method based on multilayer feature extraction of convolutional neural network. By inputting images into convolutional neural network for training, this method implements feature extraction at each full connection layer of network, inputs the features obtained into classifier, and then compares the output results. The lower full connection layer activated by relu function is selected as feature extraction layer, whose recognition rate is higher than that in higher full connection layer. This paper builds up office supplies dataset, and realizes the office supplies identification system based on the multilayer feature extraction of convolutional neural network. The layer relu6 of AlexNet is selected feature extraction layer, and the optimal training image quantity as well as the optimal classifier construction system is chosen, which verifies the feasibility of this method.